[cfe-commits] [PATCH] CFG automatic object destructors

[Ted Kremenek]

Not sure about this one, but it might make sense to make ReturnStmt a terminator anyway.  There's a couple issues to think about here, including how we will model CallEnter/CallReturns for inter-procedural analysis in the presence of dtors.  My initial thought is the ReturnStmt should be the site of the "jump" back to the caller, which means that dtors need to be in the cfg before the ReturnStmt itself.  It's not hard to modify the analysis engine (in a separate patch) to have ReturnStmt be a terminator if that is the right thing to do.

On Sep 27, 2010, at 10:18 PM, Zhongxing Xu <xuzhongxing at gmail.com> wrote:

> I might be wrong. Shall we add the implicit dtor before or after the return stmt? I guess we should add them after the return stmt, since object may be returned. But that means we would have CFGElements after the terminator.
> 
> 2010/9/28 Zhongxing Xu <xuzhongxing at gmail.com>
> Hi Marcin,
> 
> There are some bugs in your patch. I tested with the code attached. The printed CFG has some errors in block B2.
> 
> I'll look at the patch to try to locate the bug.
> 
> 2010/9/28 Marcin Świderski <marcin.sfider at gmail.com>
> 
> Patch adds implicit destructors generation for objects with automatic storage duration. Patch is rather big, but it contains all cases (that I thought of), as it will be easier to review as a whole IMO. While commiting I can divide it to smaller chuncks.
> 
> I've added some test cases. They can be later used to create regresion tests.
> 
> Please approve for commit. 
> 
> Marcin
> 
> 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20100928/9bc1853a/attachment.html>